Research On Feature Point Detection Algorithm Of Polarization Image Of Metal Fatigue Damage Surface Based On KAZE

The fatigue damage of metal components is complicated.Most of the components are not disassembled during use.Traditional mechanical analysis and grain analysis are difficult to monitor the fatigue damage of metal in real time.In fact,in the process of fatigue damage of metal components,the surface texture of the components is constantly changing,and the polarization characteristics of the metal components with different degrees of fatigue damage are significantly different.the edge,texture and detail features of the object can be highlighting by contrast the difference of the polarization characteristics on the surface.Then,the polarization information such as polarization degree and polarization angle on the surface of the object can be calculate by polarization images with different polarization azimuth angles,and the surface of the metal component can be analyzed and simulated.The mathematical relationship between polarization information and fatigue damage degree can be analyzed and simulated by the information.Thus the fatigue damage degree of the metal can be predicted by the surface polarization characteristics.Differences in displacement,viewing angle and illumination between polarized images with different polarization azimuths are caused by inconsistency of each channel.In order to accurately calculate the polarization information,it is necessary to register polarization images with different polarization azimuths.The key step in registration is to find the exact correspondence between images by feature detection algorithm.The traditional SIFT and SURF feature detection algorithms can adapt to the rotation,angle of view,scale and other transformations between images.But the feature space is constructed by linear diffusion filtering.The edges and details of the image are lost,and the correspondence between the images cannot be accurately found.The KAZE algorithm uses the anisotropic thermal diffusion equation to construct a nonlinear scale space,and preserves the edge and detail features of the image during the filtering process.Based on this,this paper introduces the method of polarized light detection for the detection of surface fatigue damage characteristics of metal structural parts,focusing on the extraction of feature points that affect the registration accuracy of polarized images.The main contents include the following three aspects:1)During the fatigue damage of metal components,the surface texture changes from regular to irregular,changing the polarization characteristics of the metal surface.Through experimental analysis,it is concluded that during the fatigue damage of metal components,the surface first has regular grain slip.As the fatigue cycle increases,cracks appear on the surface.From the information entropy of solving the polarization degree image,it can be seen that the fatigue cycle has a certain correspondence with the polarization characteristics of the metal surface,which provides a possibility to predict the degree of metal fatigue damage through the polarization characteristics.However,the polarization detection images with different polarization azimuths need to find the exact correspondence between the polarization detection images for registration.Otherwise,the false polarization information is easy to be resolved.The traditional SIFT and SURF algorithms are difficult to accurately detect the edge features of the polarization image.2)KAZE algorithm uses the anisotropic thermal diffusion equation to construct the nonlinear scale space,and preserves the edge and detail features of the image during the filtering process.However,the nonlinear equations in the PM model have no solution and diffusion filtering speed,resulting in image weak edge loss.In order to solve this problem,an improved image feature point detection algorithm(CKAZE)for KAZE is proposed.In this algorithm,an adaptive diffusion filter function is constructed based on KAZE algorithm and energy functional principle.This function can make the solution of nonlinear diffusion equation unique and preserve the weak edge features of the image,which improves the feature.The point describes the progress of the vector,which reduces the mismatch rate of the feature points.Finally,the Mikolajczyk standard database and the polarization image of the metal fatigue damage process are used to perform the feature matching experiment.From the experimental results,the nonlinear diffusion solution of the CKAZE algorithm is unique.The weak edge features of the image in the linear scale space can be well preserved,and the correct logarithm of the feature matching is relatively improved.3)The KAZE algorithm has a reduced image detection capability for a single structure and is sensitive to image brightness and contrast changes,so it cannot adapt to the complexity of the environment.Based on this,a feature point detection algorithm(NPI)for nonlinear polarization images is proposed.The algorithm uses nonlinear diffusion filtering to construct a nonlinear scale space.The weak edge detection ability is strong.The Hessian determinant and the strong edge detection ability are strong.The ratio is used as a criterion for the feature points.Finally,the point and line images,the Mikolajczyk standard database image and the polarization image of the metal surface in some complex environments are used for feature detection.From the experimental results,the algorithm can still have a very simple image anda partially complex environment.Strong feature detection capability provides a solid foundation for studying fatigue damage through polarized images.